The Public Speaking Final Exam: Studying for a Test About Talking

Every skill you have ever learned splits into two parts: the part you can be told, and the part you can only get by doing it badly a number of times. Riding a bike. Interviewing someone. Writing a good bug report. The two halves are not equally teachable, and — this is the part that matters — they are not equally testable.

A public speaking course puts that split under a spotlight, because its final exam has to be written.

The instructions, verbatim

Before anything else, the acceptance criteria. The exam opened with these:

This exam includes multiple-choice, true/false, and short essay questions. Once you begin the exam, you must complete it in one sitting. You will not be able to exit and return later.

The exam has a three-hour time limit… Any answers not submitted by that time will not be graded. No extensions or exceptions will be granted.

This is an open-book exam. You may reference your class notes and course materials. However: Internet searches and use of AI tools are not permitted.

Essay question responses must be between 150 and 200 words. Responses that exceed or fall short of this range will result in a half-point deduction.

That last one is my favourite line in the whole document, and I want to defend it before I take anything apart.

A word range with a stated penalty is a real acceptance criterion. It is observable, it is symmetrical — you are punished for padding and for skimping — and it makes brevity a graded skill rather than a personal preference. In a course whose central discipline is fitting an argument into a fixed time, examining the written answers under a fixed length is the closest a page can get to a timer.

Then the exam pointed me at two recorded talks and asked fifty-four questions. Not one of them asked me to speak.

What the fifty-four questions were actually measuring

I sorted them, because sorting things is how I calm down. The result was not what I expected going in.

I went in assuming the exam was a diagnosis test — two recordings, and questions about what the speakers did. That is the smaller half of it. Fifty-five per cent of the items, and just under half the points, are course doctrine that never touches either talk: name the steps, define the term, say which pattern takes a preview statement. The recordings supply nineteen items and three of the five essays. Everything else is vocabulary.

That isn’t damning, but it is worth stating plainly, because the exam is not the whole assessment — the graded speeches are, and those happened live in front of people. This paper was the part that could be examined as knowledge, and someone made a deliberate choice about what counted.

The choice they made was: the knowledge half of speaking is mostly vocabulary, with diagnosis on top. Can you name the parts, and then can you watch a competent person and say which ones they used?

I think that’s the right call, and I’d defend it in a design review — the ordering especially. The vocabulary is what turns a vague impression into an actionable note. “That talk dragged” is a feeling. “The central claim doesn’t land until 7:37 and never gets restated” is a fix. You cannot produce the second sentence without the words from the first thirty questions.

The essay I’d defend, and the one that gives me pause

Five questions were worth five points each, and they were the only places the exam let me build an argument rather than pick one.

One of them asked whether a speaker’s organisation was effective, and how I would improve it, with timestamps. My answer, at 172 words, said yes — and then this:

It takes nearly seven minutes to reach the idea that addiction may stem from our need to bond and connect, and about seven more to conclude that when close relationships or engaging to-do lists are missing, we may seek bonding with substances. Although I enjoyed his speech, I would surface the ending statement earlier and repeat it between examples.

That’s a defensible note, with the evidence attached: the thesis at 7:37, the closing line at 14:27, and roughly seven minutes of groundwork before either.

It is also — I noticed this only afterwards — a word-for-word application of the rule the course had drilled all term. The topical pattern’s whole doctrine is preview it, say it, review it: repeat, repeat, repeat. Asked to improve a professional’s talk, what I produced was the course’s own instruction, aimed outward.

I don’t think that’s cheating. It’s evidence the teaching took. But it is a narrow kind of evidence. The exam confirmed that I can apply the rule to a recording I can pause. It has nothing to say about whether I apply it standing up, four minutes on the clock, with an outline I’m trying not to look at.

Where the instrument slips

Two of the true/false items ask, in different words, whether the first speaker’s talk was informative rather than persuasive.

The talk in question is a man arguing that adults should let children do things adults find frightening. He builds a need, mocks the opposing view, and closes with a list of actions. It is shaped like an informative talk and it is doing persuasive work — which is the interesting thing about it, and precisely what makes it worth showing to a class.

A true/false item cannot hold that. It has to pick. And when a genuinely mixed case gets forced onto a binary, what the item ends up measuring is not “do you understand the informative/persuasive distinction” but “did you attend the session where this specific talk was labelled.” That’s a different construct wearing the same clothes — the same failure mode I ran into on the measurement side of things, where a score quietly reports something other than what it claims to.

It’s four points out of a hundred and twenty-three. I’m not building a case; I’m noting that the place the instrument gets shaky is the place the underlying idea is genuinely contested, and that this is where instruments usually get shaky.

There is a larger version of the same thing, and it is worth more marks. Two of the five essays cannot be answered by anyone who understands rhetoric perfectly but wasn’t in the room — one turns on the speeches classmates gave over the term, the other on the shared vocabulary that accumulates in any long-running course. Ten points, scored on presence.

I don’t think that’s a defect, and I’d probably keep both. A course is a room, participation is real, and rewarding it is a legitimate choice. But it should be counted honestly: those ten points measure attendance and belonging, not speaking. Averaged into a single score labelled public speaking, they quietly make the number mean something other than it says — which is the same failure as the true/false pair, just larger and better disguised, because nobody suspects an essay.

The split, stated plainly

Sorted by what an instrument on paper can actually reach, the fifty-four items fall into three tiers — and the exam’s weighting follows them almost exactly, heaviest where the paper is strongest.

The exam knows this about itself. That’s why it exists alongside the speeches rather than instead of them.

What I took from it

I have spent my working life on the other side of this problem: writing the check that decides whether something passes, and living with the gap between what the check observes and what I actually care about. It is very easy to gate a release on a coverage number, watch the regression suite go green in headless silence, and let that stand in for stability. The suite proves every line executed. It has nothing to say about whether the thing stutters under load, or falls apart in the hands of someone using it for the first time.

This exam doesn’t do that, and the reason is structural rather than clever. There are two instruments. The written one measures the vocabulary and says so. The live one measures the performance and can’t be faked. Neither pretends to be the other, and nobody averaged them into a single figure that would have meant nothing.

That’s the transferable part, and it has nothing to do with speaking: when a skill has an untestable half, the fix is a second instrument, not a broader claim about the first one. I have shipped the broader claim more than once.

Thanks to an instructor who set a word range with a penalty attached and made me feel it at 158 words. If you assess anyone at work — interviews, reviews, certifications — which half of the thing you care about is your instrument actually reaching? I’d like to hear from someone who has built the second one.